With the advent of semiconductor devices having increasingly large component densities, the removal of heat generated by the devices has become an increasingly challenging technical issue. Over time, the frequency of operation of CMOS devices has increased significantly. The resulting microprocessor power dissipation has likewise gone up by an order of magnitude. While the input voltage and capacitance of devices have decreased, the number of devices on a typical microprocessor die continues to increase rapidly as processing efficiency is sought. Moreover, device miniaturization has led device designers to integrate previously separate components, such as those used to create a cache, into the microprocessor die.
This consolidation of devices has resulted in high CPU core power density, e.g., 50% of a 20 mm by 20 mm microprocessor die may contain the CPU core, with the rest being cache. Furthermore, typical processor boards can, in some instances, include multiple CPU modules, application-specific integrated circuits (ICs), and static random access memory (SRAM), as well as a dc-dc converter, all of which have increasing power dissipation requirements, thereby increasing the total power dissipation level needed by computer systems.
Air-cooled heat sinks can be used to increase the heat-dissipating surface area of heat-producing devices. High-conductivity gases have been used within such heat sinks to increase the efficiency of the heat transfer within the divice. However, heat sinks are typically characterized by a mechanical interface to their cooled devices, which commonly leads to interference in the heat flow, and can lead to very high thermal resistance. Indeed, the bulk of the available thermal budget for cooling, typically a 45 degrees C. temperature differential between the chip temperature and the ambient temperature, will commonly be used up by this interface. The mechanical interface can also lead to uneven cooling. This is further complicated by the non-uniform power distribution on many chips, which often results when different components are integrated onto a single chip.
To deal with these difficulties, innovative ways have been developed to reduce chip-to-heat sink thermal resistance. Included among the cooling methods for semiconductors are free-flowing and forced-air convection, free-flowing and forced-liquid convection, pool boiling (i.e., boiling a liquid cooling fluid off a submerged device), and spray cooling (i.e., boiling a liquid cooling fluid off a device being sprayed with the liquid). Because liquids typically have a high latent heat of vaporization, these latter two methods provide for a high heat-transfer efficiency, absorbing a large quantity of heat at a constant temperature.
The use of these boiling/vaporizing methods is limited to a maximum power density, the critical heat flux (CHF). At higher densities, the vaporized cooling fluid forms a vapor barrier insulating the device from the liquid cooling fluid, thus allowing the wall temperature of the device to increase greatly. This phenomenon is referred to as dry-out. When a coolant is properly sprayed, it can disperse such a vapor layer, and its CHF can be well over an order of magnitude higher than the CHF of a pool-boiling system. This high CHF is preferably a uniform spray, and should match the power dissipation requirements of the device. Thus, spray cooling presently provides the most efficient cooling for a heat-generating device, such as a semiconductor device.
In a typical spray cooling system, a cooling fluid is sprayed through an evacuated chamber at the component. Typically, the cooling fluid used for spray cooling has a relatively low boiling point (in relation to the operating temperature of the device), which is the temperature that the sprayed device is cooled toward. Most preferably, the cooling fluid is inert to the heat source. For semiconductor devices, low boiling point fluids such as 3M® FC-72, (FC-72, i.e., FLUORINERT®, sold by 3M® Corporation), 3M's Novec line of fluids (HFE 7100, etc., sold by 3M® Corporation) or PF-5060 are among a number of known suitable cooling liquids. Water may also be preferable in some instances.
A number of factors affect the performance of spray cooling, thus affecting the heat transfer coefficient h and/or the CHF. It is commonly understood that surface roughness and wettability of the sprayed component are two of these factors, and the orientation of the surface being sprayed can be a third. In particular, it is believed that h is higher for rough surfaces when using a pressurized liquid spray, and for smooth surfaces when using gas atomizing. Surfaces with decreased wettability appear to have a marginal increase in h.
Critical to consistent, controlled cooling is the controlled application of the liquid cooling fluid in a desired distribution, flow rate, and velocity. For example, at a low mass flow rate, CHF and h increase with the mass flow rate. However, at a critical mass flow rate, the advantages of increased mass flow are diminished due to pooling and/or due to a transition to single phase heat transfer. Thus, a spray cooling system is preferably operated uniformly at a mass flow rate defined at a point before the critical mass flow rate is reached. All of these factors make critical the design of the sprayer, i.e., the design of the nozzle and its related spray devices.
Also important is the interaction between sprayed cooling fluid and the surface of the component. Splashing of the cooling fluid can cause significant deterioration of the heat transfer performance. This can result in hot spots being formed at low power density regions adjacent to the regions covered by the spray. Additionally, not all portions of a component are spray cooled, thus allowing for hot spots in lower heat flux areas, and in large thermal gradients.
Also important to the cooling system design is its operating temperature. In particular, it is desirable to configure the system to operate at a high h, which will occur with a design temperature above the boiling temperature and below a temperature that will dry out the sprayed coolant. The amount of heat to be dissipated must be less than the CHF.
Accordingly, there has existed a need for a small, accurate, reliable and cost-efficient spray cooling system that can be used to support complex processing systems having one or more high-dissipation devices. The spray cooling system will preferably offer efficient power usage and an effective life span from the standpoint of a computer system operator. Preferred embodiments of the present invention satisfy these and other needs, and provide further related advantages.